Page 243 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

220 Chapter 9. Error-Resilience Video Coding Techniques

primary role. Thus, the decoder attempts to conceal the e ects of errors by
providing a subjectively acceptable approximation to the original data. This is
achieved by exploiting the limitations of the human visual system and the high
temporal and=or spatial correlation of video sequences. Error concealment is
an ill-posed problem since it does not have a unique solution. Thus, error con-
cealment techniques exploit a priori knowledge of the characteristics of video
signals to restrict the otherwise large number of possible solutions. Depending
on the information used for concealment, postprocessing techniques can be
divided into three main categories: spatial techniques, temporal techniques,
and hybrid techniques.

9.7.1 Spatial Error Concealment
Spatial techniques exploit the high spatial correlation of video signals and
conceal damaged pels in a frame using information from correctly received
and=or previously concealed neighboring pels within the same frame. Such
techniques apply primarily to intracoded blocks but may also be used to con-
ceal intercoded blocks with missing motion information or to recover the DFD
signal.
In Ref. 190 a damaged pel within a block is interpolated from the four
corner pels outside the block, as illustrated in Figure 9.7(a). Interpolation
from the four nearest pels outside the block boundaries, as illustrated in
Figure 9.7(b), is proposed in Ref. 191. Interpolation in the frequency do-
main has also been used. For example, in Ref. 192 the DC coe$cient of
a damaged block is recovered as the average or the median of the DC

f (x , y 1 ) f b (x , y 1 ) f T (x, y 1 )
1
a
2
,
f ( y )
x
d T
f , y )
(
x
f L (x 1 , ) y d d f R (x , y)
2
L R
d B
2 2
f c ( 1 x , y ) f d ( 2 x , y ) f B ( x , y 2 )
f ( x , ) y = 1 ( − x n )( − ) f a ( 1 x , y ) + x n 1 ( − y n ) f b (x , y ) 1
1 2 1 f ( x , ) y =
1 y n
+ 1 ( − x n )y n f c ( 1 x , y 2 ) + x n y n f d (x 2 , y 2 ) d L + d R + d T + d B
×[d L f R ( x , y) + d R f L ( x , y) + d T f B (x, y ) + d B f T ( y x , 1 )]
2
1
2
where x = x − x 1 , y = y − y 1
n
n
x 2 − x 1 y − y 1
2
(a) using corner pels (b) using nearest neighboring pels
Figure 9.7: Error concealment using spatial interpolation